How we benchmarked seven web-search APIs.

Definition. The share of questions where at least one of the ten returned URLs matches one of SimpleQA's gold reference URLs for that question. Binary per question, averaged over 1,000.

Normalization. Before comparison both sides are canonicalized: fragments stripped, tracking parameters removed (utm_*, fbclid, gclid…), host lowercased, www. and m. prefixes dropped, Wikipedia language subdomains collapsed, trailing slashes removed. Meaningful query parameters are preserved, so two distinct pages on the same path are not merged.

How to read it. This asks whether an engine surfaces the specific source a human researcher cited for that fact.

It is a floor, not a ceiling. SimpleQA's reference list is not exhaustive, so a provider that returns a different but perfectly good source scores zero on that question. The true “found a usable source” rate is higher than these numbers for every provider. We report it because it is the metric that actually discriminates: a same-domain variant puts all seven within 2.2 points of each other (94.8–97.0%) — everyone can find the right site, and the interesting differences are about which page gets surfaced.

Confidence intervals are Wilson score intervals at 95%. Providers whose intervals overlap are not distinguishable at this sample size.

Definition. Mean wall-clock time from request sent to response received, measured with exactly one request in flight globally, over 150 questions — 1,050 sequential calls.

Why it is measured in a separate pass. Timings collected during the main concurrent run are not provider latency. We verified this with a back-to-back A/B: the same providers measured concurrently and then immediately at concurrency 1 differed by 1.5× to 7.2×, and the inflation was not uniform — it tracked the per-provider concurrency we had assigned to respect each vendor's rate limits. That reordered the ranking: the provider that looked fastest under load was second slowest in isolation, and vice versa. In-run timings rank our own worker allocation, not vendor speed, so they are not used.

Providers are visited round-robin per question, in shuffled order, so no provider is systematically measured first and any drift in network conditions is shared evenly.

Caveat. Measured from a single location on a single day. Treat these as relative comparisons under identical conditions, not as absolute figures you will reproduce from a different network.

Definition. Vendor list price in USD per 1,000 search requests, verified 2026-09-03. All seven are billed per request, not per token.

Validation against observed spend. Two providers return billing data in the response body, which lets us check list price against reality:

• Exa reports costDollars on every call. Measured across all 1,000 calls: $7.00 per 1,000 — matching its advertised rate exactly.
• Linkup was checked by differencing its credit balance before and after. Observed $5.63 per 1,000 against an advertised $5.00. The gap is our own retry behaviour — retried requests are billed but recorded once — not a pricing discrepancy.

Since announced prices held up, announced prices are what we publish.

One figure is weaker than the others. Serper's public pricing page was not reachable and its credit packs sit behind signup, so its rate comes from third-party sources rather than a live vendor page. It is also the number that most changes the comparison, since it would make Serper the cheapest per call by a wide margin. Treat it as indicative and verify against your own account.

Price depends on commitment for some vendors: Tavily ranges from about $5.00 at high volume to $8.00 pay-as-you-go, and Keenable's 100,000 free requests per month mean this entire study cost nothing there.